We examine the investment skill of socially responsible investment (SRI) fund managers. Prior studies use the 'alpha' from standard asset pricing models as a proxy for management skill. However, implicit in the use of such models is that managers operate under no investment constraints. In the SRI context, this is patently false and can lead to biased alpha estimates and false conclusions about the existence of skill. We introduce a novel three-factor Fama-French asset-pricing model with the aim of estimating alpha more accurately and hence investment skill, without bias. This model excludes SRI-prohibited industries such as defense, alcohol, tobacco and gambling in the construction of the Fama-French market, value and size risk factors. We show that the exclusion of the SRI-prohibited industries leads to subtle and complex changes to the risk factors that drive SRI returns. When we re-estimate alpha using the new model we find, in contrast to the conventional Fama-French model, evidence of statistically and economically significant alpha. Furthermore, the risk loadings on the new risk factors are similar to those of the original Fama-French model suggesting that changes in risk loadings are not responsible for the finding of significant skill. © 2012 Springer Science+Business Media B.V.